Multi-GPU CUDA stress test(support int8 and float16 compare) http://wili.cc/blog/gpu-burn.html
git clone https://github.com/zaykl/gpu-burn
cd gpu-burn
sudo docker build -t gpu_burn .
#burn with all gpu
sudo docker run --rm --gpus all gpu_burn
#burn with single gpu
sudo docker run --rm --gpus "device=0" gpu_burn
#interact with gpu_burn
sudo docker run -it --gpus all gpu_burn /bin/bash
#load with local data
sudo docker run -it -v ./:/data --gpus "device=0" gpu_burn /bin/bash
#compare float32/float16/int8 gemm
make test_gemm
make test_int
./test_gemm
./test_int
To build GPU Burn:
make
To remove artifacts built by GPU Burn:
make clean
GPU Burn builds with a default Compute Capability of 5.0. To override this with a different value:
make COMPUTE=<compute capability value>
CFLAGS can be added when invoking make to add to the default list of compiler flags:
make CFLAGS=-Wall
LDFLAGS can be added when invoking make to add to the default list of linker flags:
make LDFLAGS=-lmylib
NVCCFLAGS can be added when invoking make to add to the default list of nvcc flags:
make NVCCFLAGS=-ccbin <path to host compiler>
CUDAPATH can be added to point to a non standard install or specific version of the cuda toolkit (default is /usr/local/cuda):
make CUDAPATH=/usr/local/cuda-<version>
CCPATH can be specified to point to a specific gcc (default is /usr/bin):
make CCPATH=/usr/local/bin
CUDA_VERSION and IMAGE_DISTRO can be used to override the base
images used when building the Docker image target, while IMAGE_NAME
can be set to change the resulting image tag:
make IMAGE_NAME=myregistry.private.com/gpu-burn CUDA_VERSION=12.0.1 IMAGE_DISTRO=ubuntu22.04 image
GPU Burn
Usage: gpu_burn [OPTIONS] [TIME]
-m X Use X MB of memory
-m N% Use N% of the available GPU memory
-d Use int8
-tc Try to use Tensor cores (if available)
-l List all GPUs in the system
-i N Execute only on GPU N
-h Show this help message
Example:
//burn fp16 600 second
./gpu_burn 600
//burn int8 600 second
./gpu_burn -d 600